U.S.-China Humanoid Race: Tesla Prepares for Mass Production as Chinese Companies Accelerate Production and Exports

finance.biggo.jp reported that Tesla is preparing for mass production of Optimus, a humanoid robot designed to work like a person, while Google DeepMind and Nvidia are also building out AI and computing and training infrastructure for robots. The article describes the U.S. as seeking an early lead in the “brains” and platforms, while Chinese companies are stepping up their push in production and exports. On the Chinese side, it cites Unitree, AgiBot and UBTECH, saying AgiBot has produced 15,000 robots cumulatively and Unitree has launched the G1 starting at 99,000 yuan and the R1 starting at 39,900 yuan.

Tesla's preparations for mass production, Google DeepMind's VLA model and Nvidia's training infrastructure show that competition is open not only to companies that make humanoids but also to those that control the environments where they are trained. Chinese companies' production records and pricing show that the supply of physical robots is a prerequisite for gathering data and improving performance, a gap that may be difficult to close through research capabilities alone.

The article's central point is that the humanoid race is beginning to split between the stages of “research announcements” and “preparations for mass production” and “production in volume.” Tesla has signaled that it is installing an Optimus production line and preparing for mass production this year, while Google DeepMind and Nvidia are seeking to establish platforms for robots' intelligence, training and validation. By contrast, Chinese companies are emphasizing actual shipments, deployment and data accumulation, citing AgiBot's cumulative production of 15,000 units, Unitree's price cuts and UBTECH's field testing. In other words, the focus of competition appears to be shifting from showcasing the performance of individual robots to how quickly companies can increase the number of physical robots available for training. A related story from this newspaper, Changying Precision ships more than 1.1 million precision parts for humanoid robots from January to August 2026, showed that parts for humanoids are already being discussed in terms of volume. This article reinforces the picture of end-product assembly and downstream parts supply starting to move at the same time. While Tesla is heading toward mass production based on its own manufacturing capacity, the article emphasizes that Chinese companies use nearby supply chains to source motors, reducers, sensors and batteries, and to accelerate design revisions. This suggests that competition is entering a phase in which success depends not on software alone but on the ability to link physical robots, parts, factories and field data. Many questions remain unresolved. The article does not specify Tesla's actual annual production volume or utilization rate, or how much of the process it will handle in-house. It is also not yet clear which robot models Google DeepMind's VLA model will be applied to, or to what extent. For Nvidia's Cosmos and Isaac GR00T, the key question is how many companies will adopt them as standard platforms. And while the article gives cumulative production figures for Chinese companies, it does not establish how much their ongoing shipments and field data contribute to improving training.

According to the article, Tesla said in its first-quarter earnings materials this year that it was installing a production line for Optimus. It is also reportedly preparing for mass production this year. The article describes Google DeepMind's “Gemini Robotics 2,” released last month, as a VLA model that understands both human language and visual information from cameras and controls the torso and legs as well as the hands and arms. It says Nvidia is building a development ecosystem through “Cosmos” and “Isaac GR00T” that trains and validates robots in virtual environments before applying them to physical robots.

The article makes no direct mention of Japanese companies or the Japanese market. However, if humanoid competition is driven by a cycle of mass production, field deployment and retraining on data, Japanese companies with strengths in parts supply and manufacturing processes may find competition in core components a more immediate focus than competition in finished robots.